Versatile PII Redactor

Inspiration

As avid TikTok users, we've witnessed countless creators facing anxiety over potential content violations and, more critically, the accidental exposure of personal information like home addresses, license plates, or identification documents.

These privacy breaches can have serious real-world consequences, from doxxing to identity theft. We were inspired to create a proactive solution that empowers creators to share content confidently while maintaining their privacy and safety.


What it does

Our Versatile PII Redactor is an intelligent privacy protection tool that automatically scans video content for personally identifiable information (PII) before publication.

The system:

  • Detects sensitive information including:

    • πŸ‘€ Human faces and identifying features
    • πŸš— License plates and vehicle identification
    • πŸͺͺ Government-issued IDs and documents
    • 🏠 Street addresses and location markers
    • πŸ’³ Credit cards and financial information
  • Applies intelligent redaction through blur effects, pixelation, or masking

  • Processes content in real-time to maintain creator workflow efficiency

  • Maintains video quality while ensuring privacy compliance

The tool acts as a comprehensive safety net, enabling creators to focus on content creation without privacy concerns.


How we built it

Our technical stack leverages cutting-edge machine learning and video processing technologies:

  • Core ML Engine: Integrated Hugging Face Gilbert ML model for robust PII detection
  • Video Processing Pipeline: Frame-by-frame analysis algorithms for temporal consistency
  • Detection Framework: Multi-class object detection for various PII categories
  • Redaction Engine: Adaptive masking algorithms that preserve video aesthetics
  • Workflow Integration: Automated pipelines for seamless video processing and reassembly

The architecture ensures scalable, accurate detection while maintaining processing efficiency.


Challenges we ran into

Building a reliable PII detection system presented several technical hurdles:

  • 🎞️ Temporal consistency across frames (motion, lighting changes, occlusion)
  • 🎯 Detection accuracy with low-resolution or partially visible elements
  • ⚑ Performance optimization for real-time processing
  • 🚫 False positive management (balancing sensitivity vs. over-redaction)
  • πŸŽ₯ Preserving video quality while applying privacy protection

Accomplishments that we're proud of

In just a few hours of intensive development, we achieved:

  • βœ… Functional Proof of Concept: Auto-detecting and redacting PII in video content
  • βœ… Multi-Modal Detection: Multiple PII categories simultaneously
  • βœ… Real-Time Processing: Achieved speeds suitable for deployment
  • βœ… Quality Preservation: Maintained video fidelity post-redaction
  • βœ… Scalable Architecture: Designed for future enhancements

What we learned

This project expanded our technical expertise and domain knowledge:

  • πŸ‘“ Computer Vision for Privacy: ML applied to privacy-preserving contexts
  • πŸ€— Hugging Face Ecosystem: Integration of pre-trained models into production workflows
  • 🎞️ Video Processing Challenges: Temporal consistency in redaction
  • πŸ”’ Privacy-First Design: Balancing user experience with strong protections
  • ⚑ Rapid Prototyping: Building functional AI systems under hackathon time constraints

What's next for Versatile PII Redactor

Immediate Enhancements

  • πŸ“· Extended PII Coverage: credit cards, computer screens, minors’ faces
  • πŸ“± Mobile Optimization: Lightweight real-time mobile models
  • 🎨 Advanced Redaction Options: Style-aware masking, creative redaction effects

Platform Integration

  • 🌐 Social Media APIs: TikTok, Instagram Reels, YouTube Shorts
  • 🎬 Video Editor Plugins: CapCut, Adobe Premiere, DaVinci Resolve
  • 🧩 Browser Extensions: Real-time video sharing protection

Long-term Vision

We envision Versatile PII Redactor as a foundational component of a privacy-first digital ecosystem, where creators can express freely while staying protected.

Our goal: make privacy protection seamless, invisible, and essential β€” an automatic safeguard in the digital age.


Built With

Share this project:

Updates